AI security research / articles
Analysis across cases.
130 Articles · Page 26 of 26
5 matching records on this page
From the archive
Long-running agents inherit state. The security question is who wrote that state, why the next agent trusts it, and whether shared memory can silently become an instruction channel.
Several 2026 failures sit outside the model container itself: shared package infrastructure, task ownership, browser origins, local control planes and connected apps.
Credentials, cloud tokens, identity records, payment data and production access recur across the 2026 cases. The useful question is not whether AI was involved, but what actually crossed the boundary.
OpenAI, Anthropic, AISI and Google/Irregular exposed different paths from an evaluation objective to real systems. Together they make one old assumption difficult to keep: the sandbox is only one part of the boundary.
A wiki edit, a malicious pull request and a package upload are different actions, but all turn the public internet into persistent state outside the intended task.
No matches on this page.